Make garbage collection thread safe

- You can now use _gc(malloc()) in multithreaded programs
- This change fixes a bug where fork() on NT disabled TLS
- Fixed TLS code morphing on XNU/NT, for R8-R15 registers

libc/nexgen32e/gc.S

@@ -20,8 +20,6 @@
 #include "libc/dce.h"
 #include "libc/notice.inc"
 
-#define INITIAL_CAPACITY 4
-
 	nop
 
 //	Invokes deferred function calls.
@@ -34,9 +32,12 @@
 //
 //	@param	rax,rdx,xmm0,xmm1,st0,st1 is return value
 //	@see	test/libc/runtime/gc_test.c
-__gc:	decq	__garbage(%rip)
-	mov	__garbage(%rip),%r8
-	mov	__garbage+16(%rip),%r9
+//	@threadsafe
+__gc:	mov	%fs:0,%rcx			# __get_tls()
+	mov	0x18(%rcx),%rcx			# cthread_t::garbages
+	decl	(%rcx)				# ++g->i
+	mov	(%rcx),%r8d			# r8 = g->i
+	mov	8(%rcx),%r9			# r9 = g->p
 	js	9f
 	shl	$5,%r8
 	lea	(%r9,%r8),%r8
@@ -55,25 +56,5 @@ __gc:	decq	__garbage(%rip)
 	mov	-8(%rbp),%rax
 	leave
 	ret
-9:	hlt
+9:	ud2
 	.endfn	__gc,globl,hidden
-
-	.bss
-	.align	8
-__garbage:
-	.quad	0				# garbage.i
-	.quad	0				# garbage.n
-	.quad	0				# garbage.p
-	.rept	INITIAL_CAPACITY
-	.quad	0				# garbage.p[𝑖].frame
-	.quad	0				# garbage.p[𝑖].fn
-	.quad	0				# garbage.p[𝑖].arg
-	.quad	0				# garbage.p[𝑖].ret
-	.endr
-	.endobj	__garbage,globl,hidden
-	.previous
-
-	.init.start 100,_init_garbage
-	movb	$INITIAL_CAPACITY,__garbage+8(%rip)
-	movl	$__garbage+24,__garbage+16(%rip)
-	.init.end 100,_init_garbage

libc/nexgen32e/gc.internal.h

@@ -12,13 +12,10 @@ struct Garbage {
 };
 
 struct Garbages {
-  size_t i, n;
+  int i, n;
   struct Garbage *p;
-  struct Garbage initmem[1];
 };
 
-extern struct Garbages __garbage;
-
 int64_t __gc(void);
 
 COSMOPOLITAN_C_END_

libc/nexgen32e/gclongjmp.S

@@ -28,29 +28,33 @@
 //	@param	esi is returned by setjmp() invocation (coerced nonzero)
 //	@assume	system five nexgen32e abi conformant
 //	@see	examples/ctrlc.c
+//	@threadsafe
 //	@noreturn
 _gclongjmp:
 	push	%rbp
 	mov	%rsp,%rbp
 	.profilable
-	lea	__garbage(%rip),%r12
-	mov	(%r12),%r13			# garbage.i
-	test	%r13,%r13
+	mov	%fs:0,%r12			# __get_tls()
+	mov	0x18(%r12),%r12			# cthread_t::garbages
+	test	%r12,%r12
+	jz	0f
+	movl	(%r12),%r13d			# garbages.i
+	test	%r13d,%r13d
 	jnz	.L.unwind.destructors
 0:	jmp	longjmp
 .L.unwind.destructors:
 	push	%rdi
 	push	%rsi
-	mov	16(%r12),%r14			# garbage.p
+	mov	8(%r12),%r14			# garbages.p
 	mov	(%rdi),%r15			# jmp_buf[0] is new %rsp
 	shl	$5,%r13				# log2(sizeof(struct Garbage))
 1:	sub	$32,%r13			# 𝑖--
 	js	2f
-	cmp	(%r14,%r13),%r15		# new %rsp > garbage.p[𝑖].frame
+	cmp	(%r14,%r13),%r15		# new %rsp > garbages.p[𝑖].frame
 	jbe	2f
-	mov	16(%r14,%r13),%rdi		# garbage.p[𝑖].arg
-	callq	*8(%r14,%r13)			# garbage.p[𝑖].fn
-	decq	(%r12)				# garbage.i--
+	mov	16(%r14,%r13),%rdi		# garbages.p[𝑖].arg
+	callq	*8(%r14,%r13)			# garbages.p[𝑖].fn
+	decl	(%r12)				# garbages.i--
 	jmp	1b
 2:	pop	%rsi
 	pop	%rdi

libc/nexgen32e/threaded.h

@@ -7,6 +7,7 @@ extern int __threaded;
 extern bool __tls_enabled;
 extern unsigned __tls_index;
 
+void TlsIsRequired(void);
 void *__initialize_tls(char[64]);
 void __install_tls(char[64]);